1. Technical Field of the Invention
The present invention relates to the field of integrated circuit, and more particularly to three-dimensional memory (3D-M).
2. Prior Arts
Three-dimensional memory (3D-M) is a monolithic semiconductor memory comprising a plurality of vertically stacked memory levels. It includes three-dimensional read-only memory (3D-ROM) and three-dimensional random-access memory (3D-RAM). The 3D-ROM can be further categorized into three-dimensional mask-programmed read-only memory (3D-MPROM) and three-dimensional electrically-programmable read-only memory (3D-EPROM). 3D-M may further comprise at least one of a 3D-memristor, 3D-RRAM or 3D-ReRAM (resistive random-access memory), 3D-PCM (phase-change memory), 3D-PMC (programmable metallization-cell memory), and 3D-CBRAM (conductive-bridging random-access memory).
U.S. Pat. No. 5,835,396 issued to Zhang on Nov. 3, 1998 discloses a 3D-M, more particularly a 3D-ROM. As illustrated in FIG. 1A, a 3D-M die 20 comprises a substrate-circuit level 0K and a plurality of vertically stacked memory levels 16A, 16B. The substrate-circuit level 0K comprises transistors 0t and interconnects 0i. The transistors 0t are formed in a semiconductor substrate 0. The interconnects 0i are formed above the substrate 0 and below the lowest memory level 16A. Hereinafter, the metal layers 0M1, 0M2 in the interconnects 0i are referred to as interconnect layers; the materials used in the interconnects 0i are referred to as interconnect materials, which include interconnect conductive materials and interconnect insulating materials.
Each of the memory levels (e.g. 16A, 16B) comprises a plurality of upper address lines (e.g. 2a), lower address lines (e.g. 1a) and memory cells (e.g. 5aa). It is coupled to the substrate 0 through contact vias (e.g. 1av). The memory cells could comprise diodes, transistors or other devices. Among all types of memory cells, the diode-based memory cells are of particular interest between they have the smallest size of ˜4F2, where F is the minimum feature size. Since they are generally located at the cross points between the upper and lower address lines, the diode-based memory cells form a cross-point array. Hereinafter, diode is broadly interpreted as any two-terminal device whose resistance at the read voltage is substantially lower than when the applied voltage has a magnitude smaller than or polarity opposite to that of the read voltage. In one exemplary embodiment, diode is a semiconductor diode, e.g. p-i-n silicon diode. In another exemplary embodiment, diode is a metal-oxide diode, e.g. titanium-oxide diode, nickel-oxide diode.
In FIG. 1A, the memory levels 16A, 16B form at least a 3D-M array 16, while the substrate-circuit level 0K comprises the peripheral circuit for the 3D-M array 16. A first portion of the peripheral circuit is located underneath the 3D-M array 16 and it is referred to as under-array peripheral circuit. A second portion of the peripheral circuit is located outside the 3D-M array 16 and it is referred to as outside-array peripheral circuits 18. It can be observed that the outside-array peripheral circuit 18 comprises significantly fewer back-end-of-line (BEOL) layers than the 3D-M array 16 and the space 17 above the outside-array peripheral circuits 18 is completely wasted. Hereinafter, a BEOL layer refers to the layer(s) defined by a single photolithography step during BEOL processing. In this example, the 3D-M array 16 comprises fourteen BEOL layers, including two for each interconnect layer (e.g. 0M1, 0M2) and five for each memory level (e.g. 16A, 16B). On the other hand, the outside-array peripheral circuit 18 comprises only four BEOL layers, including two for each interconnect layer (e.g. 0M1, 0M2).
U.S. Pat. No. 7,383,476 issued to Crowley et al. on Jun. 3, 2008 discloses an integrated 3D-M die, whose 3D-arrays and peripheral circuit are integrated on a same die. As is illustrated in FIG. 1B, an integrated 3D-M die 20 comprises a 3D-array region 22 and a peripheral-circuit region 28. The 3D-array region 22 comprises a plurality of 3D-M arrays (e.g. 22aa, 22ay) and their decoders (e.g. 24, 24G). These decoders include local decoders 24 and global decoders 24G. The local decoder 24 decodes address for a single 3D-M array, while the global decoder 24G decodes address to each 3D-M array.
The peripheral-circuit region 28 comprises all necessary peripheral-circuit components for a standalone integrated 3D-M die 20 to perform basic memory functions, i.e. it can directly use the voltage supply 23 provided by a user (e.g. a host device), directly read data 27 from the user and directly write data 27 to the user. It includes a read/write-voltage generator (VR/VW-generator) 21 and an address/data translator (A/D-translator) 29. The VR/VW-generator 21 provides read voltage VR and/or write (programming) voltage VW to the 3D-M array(s). The A/D-translator 29 converts address and/or data from a logical space to a physical space and vice versa. Hereinafter, the logical space is the space viewed from the perspective of a user of the 3D-M, while the physical space is the space viewed from the perspective of the 3D-M.
The VR/VW-generator 21 includes a band-gap reference generator (precision reference generator) 21B, a VR generator 21R and a charge-pump circuit 21W. Among them, the VR generator 21R generates the read voltage VR, while the charge-pump circuit 21W generates the write voltage VW (referring to U.S. Pat. No. 6,486,728, “Multi-Stage Charge-pump circuit”, issued to Kleveland on Nov. 26, 2002). The integrated 3D-M die 20 generates both read voltage and write voltage internally.
The A/D-translator 29 includes address translator and data translator. The address translator converts a logical address to a physical address and vice versa, while the data translator converts a logical data to a physical data and vice versa. Hereinafter, the logical address is the address at which data appears to reside from the perspective of the user and the physical address is the memory address that is represented on the address bus of the memory. Similarly, the logical data is the data transmitted from or received by the user and the physical data is the data that are physically stored in the memory cells. Note that the logical address/data are represented on the input/output 27 of the 3D-M die 20, while the physical address/data are represented on the internal bus 25 directly coupled to the 3D-M array region 22.
The A/D-translator 29 of FIG. 1B includes an oscillator 29O, an error checking & correction (ECC) circuit 29E, a page register/fault memory/trim-bit circuit 29P and a smart write controller 29W. The oscillator 29O provides an internal clock signal. The ECC circuit 29E detects and corrects errors while performing ECC-decoding after data are read out from the 3D-M arrays. It also performs ECC-encoding before data are written to the 3D-M arrays (referring to U.S. Pat. No. 6,591,394, “Three-Dimensional Memory Array and Method for Storing Data Bits and ECC Bits Therein” issued to Lee et al. on Jul. 8, 2003). The page register 29P serves as an intermediate storage device between the user and the 3D-M array(s), while the fault memory/trim-bit circuit 29P performs address mapping (referring to U.S. Pat. No. 8,223,525, “Page Register Outside Array and Sense Amplifier Interface”, issued to Balakrishnan et al. on Jul. 17, 2012). The smart write controller 29W collects detected errors during programming and activates the self-repair mechanism which will reprogram the data in a redundant row (referring to U.S. Pat. No. 7,219,271, “Memory Device and Method for Redundancy/Self-Repair”, issued to Kleveland et al. on May 15, 2007). The integrated 3D-M die 20 performs both address translation and data translation internally.
The VR/VW-generator 21 and A/D-translator 29 are outside-array peripheral-circuit components 18. Because they occupy a large area on the 3D-M die 20, the integrated 3D-M die 20 has a low array efficiency. The array efficiency is defined as the ratio between the total memory area (i.e. the chip area used for memory) and the total chip area. In 3D-M, the total memory area (AM) is the chip area directly underneath user-addressable bits (not counting bits a user cannot access) and can be expressed as AM=AC*CL=(4F2)*C3D-M/N, where CL is the storage capacity per memory level, AC is the area of a single memory cell, C3D-M is the total storage capacity of the 3D-M, F is the address-line pitch, and N is the total number of memory levels in the 3D-M. In the following paragraphs, two 3D-M dice are examined for their array efficiencies.
As a first example, a 3-D one-time-programmable memory (3D-OTP) is disclosed in Crowley et al. “612 Mb PROM with 8 Layers of Antifuse/Diode Cells” (referring to 2003 International Solid-State Circuits Conference, FIG. 16.4.5). This 3D-OTP die has a storage capacity of 612 Mb and comprises eight memory levels manufactured at 0.25 um node. The total memory area is 4*(0.25 um)2*612 Mb/8=16 mm2. With a total chip area of 48.3 mm2, the array efficiency of the 3D-OTP die is ˜33%.
As a second example, a 3-D resistive random-access memory (3D-ReRAM) is disclosed in Liu et al. “A 130.7 mm2 2-Layer 32 Gb ReRAM Memory Device in 24 nm Technology” (referring to 2013 International Solid-State Circuits Conference, FIG. 12.1.7). This 3D-ReRAM die has a storage capacity of 32 Gb and comprises two memory levels manufactured at 24 nm node. The total memory area is 4*(24 nm)2*32 Gb/2=36.8 mm2. With a total chip area of 130.7 mm2, the array efficiency of the 3D-ReRAM die is ˜28%.
It is a prevailing belief in the field of integrated circuit that integration lowers cost and more integration is better than less integration. However, this belief is no longer true for the 3D-M. For the integrated 3D-M 20, integrating the peripheral-circuit components 28 with the 3D-M arrays 22 actually increases cost. This is based on three major reasons. First of all, because the peripheral-circuit components 28 comprise significantly fewer BEOL layers than the 3D-M arrays 22, integration increases the overall 3D-M cost when the peripheral-circuit components 28 are forced to use the same expensive BEOL processes as the 3D-M arrays 22. Secondly, because they are forced to use the same number of interconnect layers (as few as two) as the 3D-M arrays 22, the peripheral-circuit components 28 are difficult to design, have a poor performance and occupy a large chip area. Thirdly, because the 3D-M cells (e.g. 5aa) require high-temperature processing, the interconnects 0i underneath the 3D-M cells need to use high-temperature interconnect materials. Being integrated, the peripheral-circuit components 28 are forced to use the same high-temperature interconnect materials, e.g. tungsten (W) for the interconnect conductive materials, and/or silicon oxide (SiO2) for the interconnect insulating materials. These materials degrade the performance of the peripheral circuit and in turn, degrade the overall 3D-M performance.